[DirectX] legalize usub.sat (#135288)

fixes #135285

This change implements the `usub.sat` intrinsic to perform an unsigned
saturating subtraction on the 2 arguments.
The minimum value this operation is clamp to is 0.

Author: farzonl

llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp

@@ -65,12 +65,27 @@ static bool isIntrinsicExpansion(Function &F) {
   case Intrinsic::dx_sign:
   case Intrinsic::dx_step:
   case Intrinsic::dx_radians:
+  case Intrinsic::usub_sat:
   case Intrinsic::vector_reduce_add:
   case Intrinsic::vector_reduce_fadd:
     return true;
   }
   return false;
 }
+
+static Value *expandUsubSat(CallInst *Orig) {
+  Value *A = Orig->getArgOperand(0);
+  Value *B = Orig->getArgOperand(1);
+  Type *Ty = A->getType();
+
+  IRBuilder<> Builder(Orig);
+
+  Value *Cmp = Builder.CreateICmpULT(A, B, "usub.cmp");
+  Value *Sub = Builder.CreateSub(A, B, "usub.sub");
+  Value *Zero = ConstantInt::get(Ty, 0);
+  return Builder.CreateSelect(Cmp, Zero, Sub, "usub.sat");
+}
+
 static Value *expandVecReduceAdd(CallInst *Orig, Intrinsic::ID IntrinsicId) {
   assert(IntrinsicId == Intrinsic::vector_reduce_add ||
          IntrinsicId == Intrinsic::vector_reduce_fadd);
@@ -586,6 +601,9 @@ static bool expandIntrinsic(Function &F, CallInst *Orig) {
   case Intrinsic::dx_radians:
     Result = expandRadiansIntrinsic(Orig);
     break;
+  case Intrinsic::usub_sat:
+    Result = expandUsubSat(Orig);
+    break;
   case Intrinsic::vector_reduce_add:
   case Intrinsic::vector_reduce_fadd:
     Result = expandVecReduceAdd(Orig, IntrinsicId);

llvm/test/CodeGen/DirectX/usub_sat.ll

@@ -0,0 +1,60 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S  -dxil-intrinsic-expansion -mtriple=dxil-pc-shadermodel6.3-library %s | FileCheck %s
+
+; Make sure dxil operation function calls for pow are generated.
+
+define noundef i16 @usub_sat_i16(i16 noundef %a, i16 noundef %b) {
+; CHECK-LABEL: define noundef i16 @usub_sat_i16(
+; CHECK-SAME: i16 noundef [[A:%.*]], i16 noundef [[B:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[USUB_CMP:%.*]] = icmp ult i16 [[A]], [[B]]
+; CHECK-NEXT:    [[USUB_SUB:%.*]] = sub i16 [[A]], [[B]]
+; CHECK-NEXT:    [[ELT_USUB_SAT:%.*]] = select i1 [[USUB_CMP]], i16 0, i16 [[USUB_SUB]]
+; CHECK-NEXT:    ret i16 [[ELT_USUB_SAT]]
+;
+entry:
+  %elt.usub_sat = call i16 @llvm.usub.sat.i16(i16 %a, i16 %b)
+  ret i16 %elt.usub_sat
+}
+
+define noundef i32 @usub_sat_i32(i32 noundef %a, i32 noundef %b) {
+; CHECK-LABEL: define noundef i32 @usub_sat_i32(
+; CHECK-SAME: i32 noundef [[A:%.*]], i32 noundef [[B:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[USUB_CMP:%.*]] = icmp ult i32 [[A]], [[B]]
+; CHECK-NEXT:    [[USUB_SUB:%.*]] = sub i32 [[A]], [[B]]
+; CHECK-NEXT:    [[ELT_USUB_SAT:%.*]] = select i1 [[USUB_CMP]], i32 0, i32 [[USUB_SUB]]
+; CHECK-NEXT:    ret i32 [[ELT_USUB_SAT]]
+;
+entry:
+  %elt.usub_sat = call i32 @llvm.usub.sat.i32(i32 %a, i32 %b)
+  ret i32 %elt.usub_sat
+}
+
+define noundef i64 @usub_sat_i64(i64 noundef %a, i64 noundef %b) {
+; CHECK-LABEL: define noundef i64 @usub_sat_i64(
+; CHECK-SAME: i64 noundef [[A:%.*]], i64 noundef [[B:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[USUB_CMP:%.*]] = icmp ult i64 [[A]], [[B]]
+; CHECK-NEXT:    [[USUB_SUB:%.*]] = sub i64 [[A]], [[B]]
+; CHECK-NEXT:    [[ELT_USUB_SAT:%.*]] = select i1 [[USUB_CMP]], i64 0, i64 [[USUB_SUB]]
+; CHECK-NEXT:    ret i64 [[ELT_USUB_SAT]]
+;
+entry:
+  %elt.usub_sat = call i64 @llvm.usub.sat.i64(i64 %a, i64 %b)
+  ret i64 %elt.usub_sat
+}
+
+define noundef <4 x i32> @usub_sat_vec(<4 x i32> noundef %a, <4 x i32> noundef %b) {
+; CHECK-LABEL: define noundef <4 x i32> @usub_sat_vec(
+; CHECK-SAME: <4 x i32> noundef [[A:%.*]], <4 x i32> noundef [[B:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[USUB_CMP:%.*]] = icmp ult <4 x i32> [[A]], [[B]]
+; CHECK-NEXT:    [[USUB_SUB:%.*]] = sub <4 x i32> [[A]], [[B]]
+; CHECK-NEXT:    [[ELT_USUB_SAT:%.*]] = select <4 x i1> [[USUB_CMP]], <4 x i32> zeroinitializer, <4 x i32> [[USUB_SUB]]
+; CHECK-NEXT:    ret <4 x i32> [[ELT_USUB_SAT]]
+;
+entry:
+  %elt.usub_sat = call <4 x i32> @llvm.usub.sat.v4i32(<4 x i32> %a, <4 x i32> %b)
+  ret <4 x i32> %elt.usub_sat
+}